Sparse Generalized Linear Model with L0 Approximation for Feature Selection

Fits sparse generalized linear models using an adaptive ridge approximation to an L0 penalty. Supported model families include Gaussian, logistic, Poisson, gamma, and inverse Gaussian regression. The package also provides cross-validation for selecting the penalty parameter.

l0ara

Overview

l0ara fits sparse generalized linear models using an adaptive ridge approximation to an L0 penalty.

Installation

Install the package from CRAN with:

install.packages("l0ara")

Basic usage

Fit a sparse Gaussian model:

library(l0ara)

n <- 100
p <- 40
x <- matrix(rnorm(n * p), n, p)
beta <- c(1, 0, 2, 3, rep(0, p - 4))
y <- x %*% beta + rnorm(n)

fit <- l0ara(x, y, family = "gaussian", lam = log(n))
print(fit)
coef(fit)

Select the penalty by cross-validation:

lam <- c(0.1, 0.3, 0.5)
cv_fit <- cv.l0ara(x, y, family = "gaussian", lam = lam, measure = "mse")

print(cv_fit)
coef(cv_fit)
plot(cv_fit)